@virajmishra1/bench-sdk

Observability for AI agents. One line of code → live dashboard, public profile, README badge.

npm install @virajmishra1/bench-sdk

Quickstart

import { observe } from "@virajmishra1/bench-sdk";

const agent = observe({
  apiKey: process.env.BENCH_KEY!,
  agent: "my-agent",
});

await agent.task("search", { query: "hello" }, async (task) => {
  task.log("started");
  const result = await doSearch();
  task.cost(0.003);
  return result;
});

That's the whole API. The wrapper records task.start, task.end, status, duration_ms, plus any task.log() / task.cost() calls inside.

What you get

• Live dashboard at https://bench.virajmishratakehome.workers.dev/dashboard
• Public profile at https://bench.virajmishratakehome.workers.dev/@you/your-agent
• Embeddable README badge: ![](https://bench.virajmishratakehome.workers.dev/badge/@you/your-agent.svg)

API

observe(options)

interface ObserveOptions {
  apiKey: string;
  agent: string;             // slug, e.g. "mcpify"
  displayName?: string;
  endpoint?: string;         // default: bench.virajmishratakehome.workers.dev
  flushIntervalMs?: number;  // default: 2000
  maxBatchSize?: number;     // default: 50
}

agent.task(name, input, fn)

Wraps an async function. Auto-records lifecycle.

agent.event(kind, data)

Manual event.

agent.flush()

Force-flush the buffer. Auto-called on batch full or interval. Useful before process exit.

License

MIT